SentielMKT

AI sentiment vs. crowd odds every analysis anchored permanently on 0G decentralised storage.

SentinelMkt pulls live prediction markets from Polymarket, runs a real-time web search on each question, scores the sentiment with an LLM, and surfaces the divergence the gap between what the crowd is betting and what current news actually suggests. Every analysis is written immutably to the 0G blockchain, giving it a permanent, verifiable content ID anyone can retrieve.

Features

Markets Feed

• Fetches from three Polymarket sort orders simultaneously (high-volume, newest, mid-tier at offset 300), deduplicates, and Fisher-Yates shuffles — so every page load shows a different, varied mix of worldwide predictions

• Auto-classifies each market into 7 categories (Politics, Crypto, Sports, Tech, Entertainment, Science, Other) via regex keyword rules. Classification is intentionally simple — a keyword match per category, evaluated in priority order. A v2 would use embeddings for better recall on edge cases.

• Search by keyword, sort by volume / expiry / divergence, filter by category via horizontal pill tabs

Analyze Flow

1. Click Analyze on any market card

2. Tavily searches the web for the market question (5 sources, advanced depth)

3. Groq scores sentiment 0.0–1.0 using the search results as evidence

4. Divergence = |market YES price − AI sentiment score|

5. The full snapshot (question, odds, sentiment, sources, reasoning, timestamp) is uploaded to 0G Storage

6. The card expands full-width, showing the divergence score, animated comparison bar, AI reasoning, source links, and the 0G proof section with a GlitchHash content ID reveal

Divergence Leaderboard

• On page load, the top 10 markets are auto-analyzed sequentially in the background — each analysis runs the full pipeline (Tavily → Groq → 0G upload). Sequential execution ensures each wallet transaction gets a unique nonce, avoiding REPLACEMENT_UNDERPRICED errors from parallel writes.

• A live progress banner shows N of 10 complete with an animated progress bar while analyses are running

• Ranked by divergence score with labeled bars showing Market YES: X% vs AI: X% at a glance

• Paginates 5 at a time, auto-cycles every 30 seconds with a progress bar and dot indicators

On-Chain Verification

• Every stored analysis has a permanent Content ID (rootHash) — a Merkle root of the file chunks on 0G

• The Verify on-chain button re-downloads the snapshot from 0G and confirms it matches what was analyzed, proving it was never altered

• A Share link button copies a permanent URL (/analysis/[rootHash]) anyone can open to view and verify the result independently

0G Components Used
@0gfoundation/0g-storage-ts-sdk v1.2.9

MemData

Wraps an in-memory Buffer as a 0G-compatible file object. Used to package the JSON snapshot before uploading no disk writes required.

Indexer

The primary interface to 0G Storage. Handles node discovery, chunk splitting, Merkle tree construction, and coordinating uploads/downloads across storage nodes.

Day 1 — May 1: Project Kickoff

We whiteboarded the idea. Scoped the stack. Connected to @0G_Labs Galileo testnet for the first time.

Day 2 — May 2:

Pulled live market data from @Polymarket for the first time today.The Sentinel was seeing for the first time raw, unfiltered market sentiment from thousands of traders.

Day 3 — May 3&4:

Integrated the AI sentiment engine today.

It now reads each market and forms independent view of what the crowd believes.Then it compares. Contrasts. Flags the gaps.

First signal fired. First divergence spotted between AI confidence and market odds

Day 4 — May 5:0G Storage Integration

Integrated 0G Storage for persistent agent memory — conversations are archived to the 0G Log layer and loaded back as context on new sessions

Day 5 — May 7:UI DESIGN

We built a live dashboard where you can watch AI sentiment clash with market odds in real time.

Day 6 — May 9:

It's Live.